# Is papr-embed-code-1-4b SOTA?

Pending benchmarks. Announced Oct 28, 2026; no independent results yet.

As of Nov 3, 2026, papr-embed-code-1-4b is pending benchmarks on isitsota: announced Oct 28, 2026, with no independent results on the boards we track yet.

Papr's open text embedding model for code retrieval (natural language to code, code to code, SQL), made by further training Qwen3-Embedding-4B on code-retrieval data. Papr says it is flat to worse than its base on non-code retrieval. Weights on Hugging Face under Apache-2.0, and hosted on Papr's embeddings API.

Canonical: https://isitsota.com/m/papr-embed-code-1-4b
Last updated: 2026-10-28
Tracking since: 2026-10-28

## Announced facts

| Fact | Value |
| --- | --- |
| Lab | Papr |
| Announced | Oct 28, 2026 |
| API | Live (`papr-embed-code-1-4b`) · POST https://memory.papr.ai/v1/embeddings with model papr-embed-code-1-4b |
| Open weights | Published · Hugging Face papr-ai/papr-embed-code-1-4b, fp32, Apache-2.0 (repo created Oct 19, card added Oct 21) |
| Parameters | 4B (fine-tune of Qwen3-Embedding-4B) |
| Task | Text embeddings for code retrieval; one dense vector per input, no schema input |
| Base model | Qwen/Qwen3-Embedding-4B |
| Embedding size | 2,560 dimensions; last-token pooling, L2-normalized; up to 8,192 tokens |
| Languages | English and code only |
| Training data note | Trained on the train splits of 8 of the 10 CoIR tasks and SciFact, so CoIR and SciFact are mostly in-domain, per Papr |

Tags marked unsure are not stated on the lab's own pages.

## What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

- [x] Publicly announced with a source
- [x] Usable by the public (API or weights)
- [ ] Listed on an independent board we track
- [ ] Enough results for a rank in a category
- [ ] Holds across two snapshots (crowns only)

## What the lab claims

Self-reported. Shown for context; not used in any verdict.

- Lab claim: CoIR mean nDCG@10 of 80.13 over 10 tasks (mostly in-domain, per Papr), run with stock mteb 2.10 in fp32.
- Lab claim: on held-out retrieval tasks it trails Qwen3-Embedding-4B on ArguAna, SciDocs, FiQA and TREC-COVID (87.04 vs 93.00) and gains on NFCorpus and CodeEditSearch.

## Sources

- REPO: [papr-ai/papr-embed-code-1-4b on Hugging Face](https://huggingface.co/papr-ai/papr-embed-code-1-4b)
- REPO: [Papr-ai/papr-embed-evals on GitHub](https://github.com/Papr-ai/papr-embed-evals)
- PRESS: [Shawkat Kabbara on X: papr-embed release article](https://x.com/shawkatkabbara/status/2115476518088716604) (2026-10-28)

Tracking continues until an independent board lists it.

## Related

- Lab: [Papr](https://isitsota.com/labs/papr.md)
- All pending releases: https://isitsota.com/releases.md
- JSON: https://api.isitsota.com/v1/models/papr-embed-code-1-4b